You don't need to touch the builder. Open the cloned Agent above, ask in plain English, and let it work across the systems it's connected to. ## Sharpen the answers with knowledge Upload your support playbook, tone-of-voice doc, escalation rules, and knowledge base articles. Every conversation references them — sharper output, no new setup. Upload context once. Every conversation references it. ## Make it run on its own When the same kinds of tickets keep landing, set a trigger — every billing question gets first-touch drafted, every refund request gets routed to the right team. One toggle, no code, no builder. Schedule runs, fire on ticket events, listen to webhooks. ## Ready to go further? When prompting and templates stop being enough — when you want the Agent to follow your playbook word for word and orchestrate triage across channels — pick the **Build visually** tab above. You're comfortable in product UIs. You don't want to wire up an Agent from scratch, but you do want it to follow your support playbook ticket by ticket. Open the cloned Agent in the builder and shape every piece. ## Connect the tools it needs Support Agents work best when they can read from and write to the systems your team already uses. * **Ticketing** — [Zendesk](/integrations/popular-integrations/zendesk), Intercom, or Salesforce Service Cloud for tickets and case history * **Knowledge base** — your KB system for citation-grounded responses * **CRM** — [HubSpot](/integrations/popular-integrations/hubspot) or [Salesforce](/integrations/popular-integrations/salesforce) for customer context * **Comms** — [Gmail](/integrations/popular-integrations/gmail), [Slack](/integrations/popular-integrations/slack) for follow-up and internal escalation See more about [Integrations](/integrations/introduction). ## Shape every surface Each piece below changes how the Agent behaves. Run it on a real ticket between every change — small iterations beat big rewrites. Edit the system prompt to match your tone, your escalation rules, your tier-1 boundary. Add or remove the actions the Agent can take — search the KB, query the CRM, draft replies, route tickets. Upload your KB articles, support playbook, and tone-of-voice guides. The Agent draws on these in every reply. Run the Agent automatically — on every new ticket, on tag updates, on SLA breaches. Get notified when CSAT drops, when an SLA is at risk, when an escalation pattern emerges. Persistent context per customer — past tickets, known account context, prior resolutions. ## Chain Agents into a workforce When a single Agent isn't enough — triage, then draft, then update the CRM, then notify the team — wire them on the Workforce canvas. Drag, connect, and run multi-Agent pipelines without code. ## Test before you trust Set up evals so regressions in tone or accuracy get caught before the Agent replies to a customer. Define test cases in the UI. Run them on every change. ## Ready to go further? When clicking through the builder starts to feel slow — when you'd rather have your AI coding assistant build and edit Agents for you, drop into custom Python or JS code blocks, or call Agents from your own app via the SDK — pick the **Build with AI** tab above. You already work in an AI coding environment. Clicking through a UI to edit response templates feels slow. Have your AI assistant build, shape, and orchestrate Agents directly from your editor. ## Programmatic GTM Programmatic GTM is an MCP server that connects your AI coding environment to Relevance AI. Describe what you want in natural language; your AI assistant reads and writes your Relevance workspace — system prompts, Tools, knowledge, triggers, workforces, evals. The fastest setup. Install the Relevance AI plugin and build Agents from your terminal. Connect from Claude Desktop, Cursor, VS Code, ChatGPT, OpenAI Codex, or any MCP client. Clone the skills repo so your coding assistant has built-in knowledge of Relevance patterns. The full reference for what you can do programmatically.